Details
Ask Redfin uses large language models connected to a retrieval-augmented generation (RAG) architecture — meaning the AI looks up current listing data before generating an answer, rather than relying solely on what it was trained on. The backend infrastructure uses Sendbird for chat delivery, Amazon Bedrock for the language model layer, and Amazon OpenSearch for data retrieval. It achieves 96% accuracy per Sendbird's benchmarks. When a question requires a licensed professional — such as legal advice or complex negotiation guidance — the tool offers to connect users directly to a Redfin agent. Approximately 59% of questions relate to property details, 10% request agent service, and 8% request home tours. The feature originated from a Redfin internal hackathon.
Have evidence about Redfin's AI practices? Submit a report.
Report a Sighting →